Feature selection in the pattern classification problem of digital chest radiograph segmentation

Knowledge-based method for segmentation and analysis of lung boundaries in chest X-ray images

We present a knowledge-based approach to segmentation and analysis of the lung boundaries in chest X-rays. Image edges are matched to an anatomical model of the lung boundary using parametric features. A modular system architecture was developed which incorporates the model, image processing routines, an inference engine and a blackboard. Edges associated with the lung boundary are automatically identified and abnormal features are reported. In preliminary testing on 14 images for a set of 18 detectable abnormalities, the system showed a sensitivity of 88% and a specificity of 95% when compared with assessment by an experienced radiologist.

Microscopic image analysis for quantitative measurement and feature identification of normal and cancerous colonic mucosa

The development of an automated algorithm for the categorization of normal and cancerous colon mucosa is reported. Six features based on texture analysis were studied. They were derived using the co-occurrence matrix and were angular second moment, entropy, contrast, inverse difference moment, dissimilarity, and correlation. Optical density was also studied. Forty-four normal images and 58 cancerous images from sections of the colon were analyzed. These two groups were split equally into two subgroups: one set was used for supervised training and the other to test the classification algorithm. A stepwise selection procedure showed that correlation and entropy were the features that discriminated most strongly between normal and cancerous tissue (P < 0.0001). A parametric linear-discriminate function was used to determine the classification rule. For the training set, a sensitivity and specificity of 93.1% and 81.8%, respectively, were achieved, with an overall accuracy of 88.2%. These results were confirmed with the test set, with a sensitivity and specificity of 93.1% and 86.4%, respectively, and an overall accuracy of 90.2%.

Automated lung segmentation in digital lateral chest radiographs

We are developing a fully automated computerized scheme for segmenting the lung fields in digital lateral chest radiographs. Existing computer-aided diagnostic (CAD) schemes and automated lung segmentation methods have focused exclusively on the posteroanterior view, despite the diagnostic importance of the lateral view. Information from the lateral radiograph is routinely incorporated by radiologists in their decision-making process, and thus computer analysis of lateral images may potentially add another dimension to current CAD schemes. Automated analysis of the lung fields in lateral images will necessarily require accurate segmentation. Our scheme employs an initial procedure to eliminate external and subcutaneous pixels. Global gray-level histogram analysis then allows for the identification of a range of gray-level thresholds. An iterative gray-level thresholding scheme is implemented using this range of thresholds to construct a series of binary images in which contiguous regions are identified and geometrically analyzed. Regions determined to be outside the lungs are prevented from contributing to binary images at later iterations. Adaptive local gray-level thresholding is applied along the initial contour that results from the global thresholding procedure to extend the contour closer to the true lung borders. This local thresholding method uses regions of interest of various dimensions, depending on the enclosed anatomy. Smoothing and polynomial curve fitting complete the segmentation. A database of 100 normal and 100 abnormal lateral images was analyzed. Quantitative comparison of computer-segmented lung regions with lung regions manually delineated by two radiologists indicated that 83% and 84% of normal and abnormal images, respectively, displayed segmentation contours within three standard deviations of the mean inter-radiologist contour degree-of-overlap value.

Automatic calculation of total lung capacity from automatically traced lung boundaries in postero-anterior and lateral digital chest radiographs

Total lung capacity (TLC) is a very important parameter in the study of pulmonary function. In the pulmonary function laboratory, it is normally obtained using plethysmography or helium dilution techniques. Several authors have developed methods of calculating the TLC using postero-anterior (PA) and lateral chest radiographs. These methods have not been often used in clinical practice. In the present work, we have developed and automated computer-based method for the calculation of TLC, by determining the pulmonary contours from digital PA and lateral radiographs of the thorax. The automatic tracing of the pulmonary borders is carried out using: (1) a group of reference lines is determined in each radiograph; (2) a family of rectangular regions of interest (ROIs) defined, which include the pulmonary borders, and in each of them the pulmonary border is identified using edge enhancement and thresholding techniques; (3) removing outlaying points from the preliminary boundary set; and (4) the pulmonary border is corrected and completed by means of interpolation, extrapolation, and arc fitting. The TLC is calculated using a computerized form of the radiographic ellipses method of Barnhard. The pulmonary borders were automatically traced in a total of 65 normal radiographs (65 PA and 65 lateral views of the same patients). Three radiologists carried out a subjective evaluation of the automatic tracing of the pulmonary borders, with a finding of no error or only one minor error in 67.7% of the PA evaluations, and in 75.9% of the laterals. Comparing the automatically traced borders with borders traced manually by an expert radiologists, we obtained a precision of 0.990 +/- 0.001 for the PA view, and 0.985 +/- 0.002 for the lateral. The values of TLC obtained by the automatic calculation described here showed a high correlation (r = 0.98) with those obtained by applying the manual Barnhard method.

Identification of lung regions in chest radiographs using Markov random field modeling

The authors present an algorithm utilizing Markov random field modeling for identifying lung regions in a digitized chest radiograph (DCR). Let x represent the classifications of each pixel in a DCR as either lung or nonlung. We model x as a realization of a spatially varying Markov random field. This model is developed utilizing spatial and textural information extracted from samples of lung and nonlung region-types in a training set of DCRs. With this model, the technique of Iterated Conditional Modes is used to determine the optimal classification of each pixel in a DCR. The algorithm's ability to identify lung regions is evaluated on a testing set of DCRs. The algorithm performs well yielding a sensitivity of 90.7% +/- 4.4%, a specificity of 97.2% +/- 2.0%, and an accuracy of 94.8% +/- 1.6%. In an attempt to gain insight into the meaning and level of the algorithm's performance numbers, the results are compared to those of some easily implemented classification algorithms.

Automated segmentation of anatomic regions in chest radiographs using an adaptive-sized hybrid neural network

The purposes of this research are to investigate the effectiveness of our novel image features for segmentation of anatomic regions such as the lungs and the mediastinum in chest radiographs and to develop an automatic computerized method for image processing. A total of 85 screening chest radiographs from Johns Hopkins University Hospital were digitized to 2 K by 2.5 K pixels with 12 bit gray scale. To reduce the amount of information, the images were smoothed and subsampled to 256 by 310 pixels with 8 bit. The determination approach consists of classifying each pixel into two anatomic classes (lung and others) on the basis of several image features: (1) relative pixel address (Rx, Ry) based on lung edges extracted through image processing using profile, (2) density normalized from lungs and mediastinum density, and (3) histogram equalized entropy. The combinations of image features were evaluated using an adaptive-sized hybrid neural network consisting of an input, a hidden, and an output layer. Fourteen images were used for the training of the neural network and the remaining 71 images for testing. Using four features of relative address (Rx, Ry), normalized density, and histogram equalized entropy, the neural networks classified lungs at 92% accuracy against test images following the same rules as for the training images.

Anatomic region-based dynamic range compression for chest radiographs using warping transformation of correlated distribution

The purpose of this paper is to investigate the effectiveness of our novel dynamic range compression (DRC) for chest radiographs. The purpose of DRC is to compress the gray scale range of the image when using narrow dynamic range viewing systems such as monitors. First, an automated segmentation method was used to detect the lung region. The combined region of mediastinum, heart, and subdiaphragm was defined based on the lung region. The correlated distributions, between a pixel value and its neighboring averaged pixel value, for the lung region and the combined region were calculated. According to the appearance of overlapping of two distributions, the warping function was decided. After pixel values were warped, the pixel value range of the lung region was compressed while preserving the detail information, because the warping function compressed the range of the averaged pixel values while preserving the pixel value range for the pixels which had had the same averaged pixel value. The performance was evaluated with our criterion function which was the contrast divided by the moment, where the contrast and the moment represent the sum of the differences between the pixel values and the averaged values of eight pixels surrounding that pixel, and the sum of the differences between the pixel values and the averaged value of all pixels in the region-of-interest, respectively. For 71 screening chest images from Johns Hopkins University Hospital (Baltimore, MD), this method improved our criterion function at 11.7% on average. The warping transformation algorithm based on the correlated distribution was effective in compressing the dynamic range while simultaneously preserving the detail information.

Time and space results of dynamic texture feature extraction in MR and CT image analysis

Texture feature extraction is a fundamental part of texture image analysis. Therefore, the reduction of its computational time and storage requirements should be an aim of continuous research. The Spatial Grey Level Dependence Method (SGLDM) is one of the most important statistical texture description methods, especially in medical image analysis. Co-occurrence matrices are employed for the implementation of this method; however, they are inefficient in terms of computational time and memory space, due to their dependency on the number of gray levels (gray-level range) in the entire image. Since texture is usually measured in a small image region, a large amount of memory is wasted while the computational time of the texture feature extraction operations is unnecessarily raised. Their inefficiency puts up barriers to the wider utilization of SGLDM in a real application environment, such as a clinical environment. In this paper, the memory space and time efficiency of a dynamic approach to texture feature extraction in SGLDM is investigated through a pilot application in the analysis of magnetic resonance (MR) and computed tomography (CT) images.

Computerized delineation and analysis of costophrenic angles in digital chest radiographs

RATIONALE AND OBJECTIVES

The authors developed a computerized method for delineating the costophrenic angles in digital posteroanterior chest radiographs to derive quantitative information that allows for detection of abnormal blunting of the costophrenic angle.

MATERIALS AND METHODS

An automated lung-segmentation scheme was used, and small regions of interest were placed in the approximate position of the costophrenic angles in 600 clinical posteroanterior chest radiographs to define a subimage for further analysis. The diaphragmatic aspect of the costophrenic angle was delineated based on column-wise contrast information, and the costal aspect was delineated based on row-wise gray-level maxima. The angle formed by the convergence of these two aspects provided the basis for assessing abnormality. Curve fitting was then performed on these segments to form a continuous costophrenic angle delineation.

RESULTS

The computer-determined angles for 1,166 hemithoraces were compared with independent diagnostic assessments by a radiologist. An encouraging level of agreement was found between these two measurements, with the area under the receiver operating characteristic curve attaining a value of 0.83.

CONCLUSION

This delineation method enhances the automated lung-segmentation scheme. Quantitative information obtained from the costophrenic angles can be used for automatic evaluation of the presence of costophrenic angle blunting, which may indicate the presence of pleural effusion.

Automated lung segmentation in digitized posteroanterior chest radiographs

RATIONALE AND OBJECTIVES

The authors developed and tested a gray-level thresholding-based approach to automated lung segmentation in digitized posteroanterior chest radiographs.

MATERIALS AND METHODS

Gray-level histogram analysis was initially performed to establish a range of thresholds for use during an iterative global gray-level thresholding technique. Local gray-level threshold analysis was then performed on the output of global thresholding. The resulting contours were subjected to several smoothing processes, including a rolling-ball technique. The final contours closely approximated the boundaries of the aerated lung regions. The method was applied to a database of 600 posteroanterior chest images. Radiologists rated the accuracy and completeness of the contours with a five-point scale.

RESULTS

Results of the subjective rating evaluation indicated that this method was accurate, with 79% of the assigned ratings reflecting moderately or highly accurate segmentation and only 8% of the ratings indicating moderately or highly inaccurate segmentation.

CONCLUSION

This gray-level thresholding-based approach provides accurate automated lung segmentation in digital posteroanterior chest radiographs.

Evaluation and progression analysis of pulmonary tuberculosis from digital chest radiographs

Pulmonary tuberculosis is one of the most common infections in the third world countries. We propose a method for automated evaluation and progression analysis of this disease by processing a sequence of digital chest radiographs. The presence of radiological features characteristic to pulmonary tuberculosis is detected within the regions of interest using suitable image processing and pattern recognition techniques. The detected features are then quantified to evaluate the extent of the disease. The measures are evaluated for a sequence of radiographs of a patient and are normalized with respect to rib size measures. Temporal changes in these normalized measures indicate the progression of the disease during the period in which the sequence of radiographs is obtained. The details of the proposed technique and results of analysis of a number of tuberculous patients are given in the paper.

Automatic background recognition and removal (ABRR) in computed radiography images

A novel method to automatically recognize and remove background signals in computed radiography (CR) images caused by X-ray collimation during projection radiographic examinations is presented. There are three major steps in this method. In the first step, a statistical curve is derived based on many hierarchical CR sample images as a first approximation to loosely separate image and background pixels. Second, signal processing methods, including specific sampling, filtering, and angle recognition, are used to determine edges between image and background pixels. Third, adaptive parameter adjustments and consistent and reliable estimation rules are used to finalize the location of edges and remove the background. In addition, this step also evaluates the reliability of the complete background removal operation. With this novel method implemented in a clinical picture archiving and communication system (PACS) at the University of California at San Francisco, we achieved 99% correct recognition of CR image background, and 91% full background removal without removing any valid image information.